Energy Economics 43 (2014) 125139

Contents lists available at ScienceDirect

Energy Economics
journal homepage: www.elsevier.com/locate/eneco

The causal relationship between renewable electricity generation and GDP growth: A study of energy sources
Adrienne Ohler , Ian Fetters
Department of Economics, Illinois State University, Normal, IL 61790-4200, USA

a r t i c l e

i n f o

a b s t r a c t
This paper examines the causal relationship between economic growth and electricity generation from renewable sources (biomass, geothermal, hydroelectric, solar, waste, and wind) across 20 OECD countries over 1990 to 2008. The results from a commonly used panel error correction model nd (a) a bidirectional relationship between aggregate renewable generation and real GDP, (b) biomass, hydroelectricity, waste, and wind energy exhibit a positive long-run relationship with GDP, (c) hydroelectricity and waste generation exhibit a short-run positive bidirectional relationship with GDP growth, and (d) biomass, hydroelectric, and waste electricity generation have the largest impact on real GDP in the long-run. We extend the analysis to consider the possibility of structural breaks and cross-sectional dependence. Accounting for cross-sectional dependence, we nd that in the short-run, increases in biomass and waste generation negatively affect GDP, while aggregate renewable and hydroelectricity increase GDP. Energy conservation policies will positively impact GDP, if the policies cause decreases in biomass or waste energy but increase hydroelectricity and wind energy. 2014 Elsevier B.V. All rights reserved.

Article history: Received 25 February 2013 Received in revised form 3 October 2013 Accepted 19 February 2014 Available online 1 March 2014 JEL classication: C3 O5 Q2 Q3 Q4 Keywords: Renewable energy Electricity generation Panel Granger-causality Biomass Waste energy Cross-sectional dependence

1. Introduction As energy costs have risen, more scrutiny has been placed on the potential negative consequences of expanded energy use; however, a reduction in energy usage could have unintended consequences for economic growth. In order to determine the impact of energy use on economic growth, a plethora of literature has looked at the relationship between energy consumption and economic growth. Payne (2010b)

provides an extensive overview of this literature, examining 101 studies over the period 1978 to 2008, but no clear consensus has been found on the causal nature of this relationship.1 From this literature, a much smaller body of work has emerged examining a possible relationship between renewable energy and economic growth. Empirical evidence on the relationship is mixed. Several studies nd a bidirectional relationship between renewable energy consumption and economic growth (Apergis and Payne, 2010a,b, 2011a,b,d, 2012a,b; Apergis et al., 2010). Sadorsky (2009, 2009b) reports no evidence of a bidirectional relationship in the short-run but nds a relationship in the long run from real GDP to renewable energy consumption. Menegaki (2011) fails to nd a bidirectional relationship, examining 27 European countries. Payne (2009, 2012) also fails to nd
1 Four primary econometric approaches are used to analyze the causal relationship: GrangerSims causality testing, EngleGranger/JohanssenJuselius cointegration and error-correction modeling, TodaYamamoto long-run causality testing, and panel cointegration error correction modeling. Of the studies examined, 23.1% showed unidirectional causality from energy consumption to GDP growth, 19.5% found causality from GDP growth to energy consumption, 28.2% show a bidirectional relationship, and 29.2% show no relationship.

The authors would like to thank James Payne, Nicholas Apergis, and Hassan Mohammadi for valuable comments and suggestions which helped improve the paper. Additionally, we would like to thank Josep Carrion-i-Silvestre and Ruhul Salim for providing the GAUSS code to run the panel stationarity test allowing for multiple breaks. Corresponding author. E-mail addresses: aohler@ilstu.edu (A. Ohler), icfette@ilstu.edu (I. Fetters).

http://dx.doi.org/10.1016/j.eneco.2014.02.009 0140-9883/ 2014 Elsevier B.V. All rights reserved.

126

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139

evidence of causality in the US. Examining energy use by sector, Bowden and Payne (2010) nd only unidirectional causality from renewable energy consumption to real GDP, similar to Pirlogea and Cicea (2012) who reports that Romanian renewable energy consumption Grangercauses output. Furthermore, the role of individual sources is important given countries' current challenges in determining the optimal mix of energy. Almost no published research exists on the nexus between individual sources of renewable energy and GDP growth. Payne (2011) examines the relationship between biomass consumption and GDP in the US, and nds a positive unidirectional relationship from biomass to GDP. Ewing et al. (2008) analyze the impact of industrial production and employment on hydroelectricity, solar, wind, wood, and waste energy. This paper seeks to contribute to the literature on the nexus between renewable energy and GDP growth by examining individual renewable sources, including biomass, geothermal, hydroelectric, solar, waste, and wind. Using electricity generation data on 20 OECD countries from 1990 to 2008, we implement a panel error correction model (ECM) to analyze the causal relationship between real GDP and each individual renewable energy source. Following Apergis and Payne (2012b), we utilize a production model framework accounting for capital and labor. The results nd evidence of a bidirectional short-run relationship between aggregate renewable electricity generation and GDP. We further test for structural breaks in the data and examine the possibility of cross-sectional dependence (CSD). Extending the panel ECM to control for CSD, we nd that renewable energy positively impacts GDP, but changes in GDP negatively impacts renewable energy. We implement a similar analysis for each individual source of energy, but an analysis of the interactions between sources is beyond the scope of this paper. Our results contribute to the literature in several important ways. First, we add to the energy-growth literature by making a distinction between electricity generation, electricity consumption, and energy consumption.2 We examine renewable electricity generation measured in MWh rather than energy consumption, because consumption refers to energy delivered to end-use sectors, and because consumption may or may not include wind and hydroelectric power depending on the data source and their measurement unit of energy. Electricity generation also differs from electricity consumption by measuring the rm's production rather than the end user's consumption. Examining electricity generation, allows us to disaggregate the data by energy source, and still follow the production framework model. The production model allows us to overcome some omitted variable bias while avoiding using ad hoc control variables. Including capital and labor in the analysis is common in the energy-economic growth and renewable energy-economic growth literature (Payne, 2010a,b). We extend the renewable energy-growth literature by examining individual sources of energy. Biomass, hydroelectricity, waste, and wind energy sources exhibit a positive long-run equilibrium relationship with GDP growth. In the short-run, hydroelectricity exhibits the largest positive Granger-causing impact on GDP growth; and GDP growth has the largest positive impact on biomass, solar, and waste energy. We further extend the analysis to examine the possibility of structural breaks and account for CSD. The results highlight the importance of controlling for CSD. While the estimates for biomass, geothermal, and solar are similar to the previous results, the results for aggregate renewable, hydroelectricity, waste, and wind change. The estimates from the CSD corrected panel ECM nd that in the short-run, waste energy has a negative impact on GDP growth, and hydroelectricity has a positive impact. GDP growth still exhibits a positive impact on biomass and waste energy, but a negative impact on hydroelectricity. Wind exhibits a positive bidirectional relationship with GDP. Finally,
2 Several studies have examined the electricity consumptionGDP relationship (Apergis and Payne, 2011; Ozturk, 2010; Payne, 2010).

geothermal exhibits a negative bidirectional relationship with GDP and solar shows a unidirectional relationship from GDP to solar energy. The results highlight the need to distinguish between renewable sources. Biomass and waste generation are important drivers in the renewable energyGDP relationship, but the environmental impacts between sources vary. For example, biomass and waste generation emit CO2, nitrogen, and sulfur into the atmosphere, despite being labeled as net neutral CO2 emitters.3 According to the EPA, US renewables averaged 1.22 lbs of SO2 per MWh and 0.06 lbs of NOx per MWh despite hydroelectricity, solar, and wind emitting negligible amounts of either. Comparatively, biomass and waste generation emit more carbon than solar, wind, and geothermal generation. Table 1 compares the average annual CO2 emissions of US power plants by primary fuel sources, including natural gas, geothermal generation, the four largest biomass fuel sources, and three types of coal generation. The average power plant emissions are reported in lbs per MWh, and demonstrate that CO2 may increase as biomass generation increases, depending on the type of biomass used and the source replaced. Municipal solid waste (MSW) averaged 2993 lbs of CO2 per MWh, emitting more CO2 than natural gas, subbituminous coal, and lignite coal. If MSW energy replaces coal, lignite, or subbituminous coal, CO2 emissions would increase.4 The remainder of the paper is organized as follows: Section 2 presents the unit root test results, examines cointegration between the variables from a production model framework, and provides estimates from a fully modied OLS model to examine the long-run relationship between renewable energy and GDP. Section 3 presents the results of a panel ECM for aggregate renewable electricity generation to test for a causal relationship between renewable energy and GDP. We then implement the same econometric model to analyze individual sources of electricity generation. Section 4 extends the previous analysis by examining the data for structural breaks and CSD. We then implement the panel ECM controlling for CSD to analyze the individual sources. Section 5 concludes with a discussion of policy implications.

2. Data, unit roots, and tests for cointegration Data were collected from the International Energy Agency's dataset on world renewable and waste energy statistics. We examine gross electricity production (GWh) by energy source for 20 countries from 1990 to 2008.5 Real GDP, gross xed capital formation, and size of the labor force were collected from the OECD. Renewable electricity generation includes biomass, hydroelectric, geothermal, solar, waste, and wind. Table 2 presents the average annual growth rate in generation for each country. Only six countries utilize geothermal and solar energy over the entire time period considered. Most countries utilize biomass, hydro, and waste energy, with biomass and waste energy contributing to most of the renewable energy growth. When utilized, wind and solar energy exhibit the largest growth rates. Comparing the GDP growth rate to the growth in renewable energy, no discernible trend appears. The top 5 countries in aggregate renewable energy growth, Denmark, Netherlands, Belgium, Portugal, and Germany, are ranked 7th, 11th, 5th, 16th, and 3rd in terms of real GDP growth.
3 Net neutral carbon emitters release CO2 as part of the natural carbon cycle of the earth. Biomass products extract CO2 from the air as they grow, and release CO2 when burned. 4 Biomass includes solid biofuels, biogasoline, biodiesels, biogases, and other liquid biofuels. Municipal waste is dened as waste collected from the residential, commercial and public service sectors, used for the production of heat and power in a central location. We examine only the renewable fraction of municipal waste. 5 Countries included: Australia, Austria, Belgium, Canada, Denmark, France, Germany, Iceland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, and United States.

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 1 Carbon emission statistics from US power plants by primary fuel source (lbs. per MWh). Source: EPA eGRID2012 Version 1.0 Plant File (Year 2009 Data) available at http://www.epa.gov/cleanenergy/energy-resources/egrid/faq.html. Coal (bit.) Total net generation (GWh in 2009) Mean Median St. dev. Minimum Maximum 861,743 3339.09 2094.15 18,130.53 686.84 323,729.63 Coal (lig.) 79,112 2423.93 2358.53 211.16 2093.42 2919.20 Coal (subbit.) 830,321 2312.50 2268.68 872.79 880.37 10,014.27 Coalbiomass (bit.) 1731 566.69 560.60 45.76 504.50 611.53 Biomass (black liquor) 23,668 184.55 134.92 162.96 0.00 851.53 Biomass (landll gas) 7606 13.05 0.00 64.13 0.00 490.16 Biomass (municipal solid) 14,450 2992.68 3320.02 1095.38 730.80 5688.32 Biomass (wood and wood waste) 16,389 112.18 0.00 278.01 0.00 2029.73 Geothermal 15,009 38.53 60.00 35.62 0.00 88.80 Natural gas 20,284

127

900.35 423.40 1050.67 29.00 3559.61

Hydro, solar, nuclear, and wind have been excluded from comparison. Their CO2 emissions output are negligible. As an extreme outlier, Southern Minnesota Beet Sugar power plant was excluded to maintain consistent and comparable results. For coalbiomass generation, the percent of generation from biomass is approximately 50% for power plants in this category.

Four potential hypotheses for the energyGDP relationship have been outlined in the literature as the growth, conservation, neutrality, and feedback hypotheses. We apply these hypotheses to renewable electricity generation such that the growth hypothesis implies causality from renewable electricity generation to economic growth, and is supported if an increase in generation positively impacts real GDP. The conservation hypothesis suggests that causality occurs from growth to electricity generation, and an increase in economic growth positively affects generation. If no relationship is found, then the neutrality hypothesis is supported. Finally, the feedback hypothesis states that renewable generation and economic growth are complementary and interdependent. This hypothesis is supported if the data conrm a bidirectional positive relationship between GDP growth and renewable generation. Applying a production function framework to the renewable energyGDP relationship, let Yit = f(Kit, Lit, REit, NREit) where Y, K, and L represent output, capital, and labor for country i in year t. Capital and labor are included in the model as a means for controlling for potential omitted variable bias (Lutkepohl, 1982). This variant of the Solow growth model considers electricity as a factor of production, separated into renewable (RE) and nonrenewable (NRE) generation. We include nonrenewable generation as a means to differentiate the impact different energy sources have on GDP. Accounting for unobserved variation and measurement errors, the linear econometric model becomes

where each variable is taken in logarithmic form, and the estimated coefcients represent elasticities. Output, Yit, measured as real GDP, is a linear function of renewable and nonrenewable electricity generation, real gross xed capital formation, and the labor force. Country xed effects and a time trend are included in the model with the parameters 1i and 2i. We extend the model to analyze six individual sources of renewable electricity generation, utilizing the production model proposed in Eq. (1). RE is replaced with each individual source, and NRE with the complement to that source. For example, the long-run econometric model for biomass is written as Yit 1i 2i t 3 REit 4 NREit 5 Kit 6 Lit it 2

where non-biomass generation is calculated as total generation minus biomass generation. We include the complement variable as a means to differentiate the impact different energy sources have on GDP, while minimizing the loss of degrees of freedom in the estimated model. The ve other renewable sources follow a similar model. 2.1. Unit root tests We implement several unit root tests for each of the variables. Tables 35 present the results for six different tests: Levin et al. (2002), Breitung (2001), Breitung and Das (2005), Im et al. (2003), two different Fisher-type tests with an augmented DickeyFuller specication and a PhillipsPerron specication (Choi, 2001; Hadri, 2000). For all tests,

Yit 1i 2i t 3 REit 4 NREit 5 Kit 6 Lit it

Table 2 Average annual growth rate from 1990 to 2008. Real GDP Australia Austria Belgium Canada Denmark France Germany Iceland Italy Japan Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States 0.063 0.042 0.041 0.049 0.042 0.035 0.032 0.081 0.046 0.006 0.074 0.051 0.052 0.073 0.066 0.071 0.045 0.029 0.053 0.052 Total RE 1.355 1.920 13.421 1.442 15.702 2.163 9.620 7.824 3.385 0.388 8.711 15.254 0.760 1.463 10.271 8.441 1.611 1.719 8.557 0.808 Biomass 10.997 8.896 24.854 4.950 17.070 3.341 27.303 41.152 2.913 24.624 3.407 2.452 4.897 11.106 10.144 5.650 18.285 0.709 Geothermal 17.980 3.163 3.350 3.853 54.204 0.190 Hydro 0.823 1.419 4.281 1.340 1.593 1.498 2.249 6.658 2.072 0.067 2.294 3.336 0.160 1.459 9.786 4.431 0.934 1.612 2.518 0.586 Solar 71.202 38.051 36.389 63.332 21.877 3.185 Waste 32.929 5.717 2.121 25.755 13.474 7.980 27.067 6.020 4.270 5.623 8.058 18.143 27.857 6.258 14.424 3.442 Wind 32.156 15.588 48.781 70.135 28.383 72.359 73.648 41.534 63.5730 18.966

128 Table 3 Panel unit root test results with aggregate renewable measure. LevinLinChu adjusted t Y Y RE RE NRE NRE K K L L
a b c

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139

Breitung 3.837 (0.999) 0.638 (0.262) 0.010 (0.4960) 6.3610a (0.000) 0.5630 (0.2870) 8.763a (0.000) 1.9150 (0.972) 1.725b (0.042) 4.256 (1.000) 3.0750a (0.001)

ImPesaranShin 1.444 (0.926) 4.874a (0.000) 0.772 (0.220) 6.416a (0.000) 0.672 (0.749) 4.848a (0.000) 1.135 (0.128) 2.452a (0.007) 0.887c (0.813) 2.487a (0.006)

FisherADF inverse 98.854a (0.000) 153.903a (0.000) 70.944a (0.002) 77.688a (0.000) 36.392 (0.633) 139.296a (0.000) 60.416b (0.020) 84.715a (0.000) 46.782 (0.214) 56.199b (0.046)

FisherPP inverse 2 p 63.957a (0.009) 139.741a (0.000) 153.525a (0.000) 530.891a (0.000) 71.208a (0.002) 371.079a (0.000) 70.890a (0.002) 156.340a (0.000) 111.578a (0.000) 270.269a (0.000)

Hadri LM z 17.048a (0.000) 2.429a (0.008) 7.274a (0.000) 2.297 (0.989) 10.205a (0.000) 0.597 (0.725) 13.832a (0.000) 3.792a (0.000) 19.275a (0.000) 5.672a (0.001)

1.437 (0.925) 4.433a (0.000) 2.414 (0.992) 5.249a (0.000) 2.067 (0.981) 2.835a (0.002) 2.383 (0.991) 3.379a (0.000) 5.025 (1.000) 5.810 (1.000)

Statistical signicance at .01 level. Statistical signicance at .05 level. Statistical signicance at .10 level.

the null hypothesis is non-stationarity, except the Hadri Lagrange Multiplier test, which assumes that all panels are trend stationary. Each test has advantages. The LevinLinChu and Breitung tests assume a common root across the countries, and Breitung has the highest power and smallest size distortion. The ImPesaranShin test allows for an individual root for each country. The null hypothesis for this test is that all panels have a unit root, and the alternative is that a fraction of the panels are stationary. The Fisher tests also allow for as much heterogeneity across countries as possible, conducting unit root tests for

each country and combining the test results. Finally, Hadri conducts a test with the null hypothesis that the data contain a unit root. For all tests, we include a country intercept and a trend. Table 3 reports the unit root test results for real GDP, capital, labor, renewable energy, and nonrenewable energy. Most of the tests conrm that all variables are non-stationary in level and stationary in rst difference. Most notably, the ImPesaranShin test result nds evidence that each variable is integrated of order one, which is the test commonly found in the renewable energyGDP literature.

Table 4 Unit root test results by renewable source. LevinLinChu adjusted t 3.897 (1.000) Bio 0.859 (0.805) Geotherm 6.415a (0.000) Geo 3.022a (0.001) Hydro 0.595 (0.724) Hydro 5.405a (0.000) Solar 4.314 (1.000) Solar 4.511a (0.000) Waste 3.192a (0.001) Waste 1.101 (0.136) Wind 1.968 (0.975) Wind 3.176a (0.001)
a b c

Breitung 1.742 (0.959) 5.388a (0.000) 0.685 (0.753) 2.960a (0.002) 1.723c (0.042) 5.644a (0.000) 4.410 (1.000) 0.810 (0.791) 2.227 (0.987) 3.079a (0.001) 2.803 (0.998) 4.925a (0.000)

ImPesaranShin FisherADF inverse 1.031 (0.849) 3.413a (0.000) 1.560b (0.059) 2.879a (0.002) 1.511b (0.065) 5.616a (0.000) 3.580 (1.000) 2.486a (0.007) 0.962 (0.168) 6.515a (0.000) 1.643 (0.950) 3.611a (0.000) 27.932 (0.830) 81.384a (0.000) 10.907 (0.537) 33.201a (0.001) 48.333 (0.172) 63.853a (0.010) 1.993 (0.999) 5.597 (0.935) 25.718 (0.776) 100.663a (0.000) 12.514 (0.897) 63.578a (0.000)

FisherPP Inverse 2 p 99.628a (0.000) 327.972a (0.000) 6.591 (0.883) 41.567a (0.000) 102.735a (0.000) 510.312a (0.000) 79.188a (0.000) 96.456a (0.000) 104.974a (0.000) 270.049a (0.000) 39.756a (0.005) 137.613a (0.000)

Hadri LM z 13.231a (0.000) 0.381 (0.352) 8.849a (0.000) 0.371 (0.356) 5.940a (0.000) 2.568 (0.995) 8.845a (0.000) 3.233a (0.001) 16.782a (0.000) 2.114c (0.017) 17.493a (0.000) 2.574a (0.005)

Countries included Australia, Austria, Belgium, Canada, Denmark, France, Germany, Italy, Japan, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States. Iceland, Italy, Japan, New Zealand, Portugal, United States.

Biomass

All

Germany, Italy, Portugal, Spain, Switzerland, United States.

Austria, Belgium, Canada, Denmark, France, Germany, Italy, Japan, Luxembourg, Netherlands, Norway, Spain, Sweden, Switzerland, United Kingdom, United States. Belgium, Denmark, Germany, Italy, Netherlands, Portugal, Spain, Sweden, United Kingdom, United States.

Statistical signicance at .01 level. Statistical signicance at .10 level. Statistical signicance at .05 level.

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 5 Unit root test results for renewable complement variables by source. LevinLinChu adjusted t Non-biomass Non-bio Non-geotherm Non-geo Non-hydro Non-hydro Non-solar Non-solar Non-waste Non-waste Non-wind Non-wind 5.551 (1.000) 13.167a (0.000) 4.628 (1.000) 1.608c (0.054) 0.1165 (0.546) 2.99a (0.001) 0.3203 (0.626) 2.410a (0.008) 1.541 (0.938) 4.454a (0.000) 1.385 (0.917) 10.045a (0.001) Breitung 0.730 (0.7673) 7.587a (0.000) 3.181 (0.999) 0.761 (0.777) 0.993 (0.161) 7.538a (0.000) 0.053 (0.521) 4.929a (0.000) 2.44 (0.596) 6.714a (0.001) 0.386 (0.650) 5.917a (0.000) ImPesaranShin 2.252 (0.988) 4.972a (0.000) 2.189 (0.986) 1.673b (0.0047) 0.297 (0.383) 4.619a (0.000) 0.642 (0.740) 1.295c (0.098) 2.215 (0.013) 7.431a (0.000) 0.833 (0.798) 2.482a (0.007) FisherADF inverse 21.060 (0.9776) 55.642b (0.0194) 2.094 (0.999) 24.135b (0.020) 53.767c (0.072) 60.507b (0.019) 1.594 (0.999) 3.948 (0.984) 27.492 (0.694) 117.310a( 0.000) 6.604 (0.998) 20.636 (0.419) FisherPP inverse 2 p 63.031a (0.0035) 388.594a (0.000) 4.952 (0.960) 98.0151a (0.000) 84.184a (0.000) 384.212a (0.000) 20.561c (0.057) 126.645a (0.000) 70.988a (0.000) 300.679a (0.000) 47.494a (0.001) 228.598a (0.000)

129

Hadri LM z 13.167a (0.000) 0.245 (0.403) 7.779a (0.000) 0.068 (0.473) 10.749a (0.000) 0.219 (0.587) 8.425a (0.000) 1.569c (0.058) 11.495a (0.000) 0.104 (0.458) 9.289a (0.000) 1.139 (0.127)

These variables are calculated in a manner similar to nonrenewable generation. For example, non-biomass generation is calculated as total generation minus biomass generation. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

Table 4 presents the unit root tests for the individual sources of energy. Depending on the test, a case could be made that each of the sources are integrated of order one. Biomass passes 4 tests, Breitung, ImPesaranShin, FisherADF, and Hadri. Geothermal and wind energy also pass 4 tests, including Breitung and FisherADF. Hydro passes 3 tests, LevinLinChu, FisherADF, and Hadri. Solar and waste pass two and three tests, respectively. The ImPesaranShin test suggests that all sources are integrated of order one with the exception of geothermal and hydroelectricity. The Hadri test suggests that these two sources are integrated of order one. We construct controls for all other electricity generation. Nonbiomass, non-geothermal, non-hydro, non-solar, non-waste, and non-wind generation account for causality from other sources. These complement variables are calculated in a manner similar to nonrenewable generation, but also contain other renewable sources. The LevinLinChu and ImPesaranShin test results demonstrate that the complement variables are unambiguously integrated of order one, as presented in Table 5.

2.2. Tests for cointegration To test for cointegration, we implement seven panel cointegration tests proposed by Pedroni (1999, 2004) which account for heterogeneous panels. The tests check for a unit root process on the residuals such that it = iit 1 + wit in Eq. (1). The null hypothesis of no cointegration, i = 1, is rejected if the cointegration term is statistically signicant. Table 6 presents the results for both panel and group tests. Panel tests require a common autoregressive coefcient, and group tests allow for individual autoregressive coefcients. Depending on the test, a case could be made that a cointegrating relationship exists between each renewable energy source, real GDP, capital, and labor. For aggregate renewable electricity generation, four of the seven tests reject the null hypothesis of no cointegration at the 5% signicance level, indicating a long-run relationship between output, capital, labor, and renewable and nonrenewable generation. In column 3, ve of the seven tests indicate a long-run relationship between output, capital, labor, biomass, and non-biomass generation. Biomass,

Table 6 Panel cointegration tests by source. Aggregate renewable Within dimension Panel v-stat Panel rho-stat Panel pp-stat Panel adf-stat Between dimension Group rho-stat Group pp-stat Group adf-stat 10.805a 3.847 0.343 1.874b 4.820 2.830a 3.151a Biomass 15.30a 3.027 2.270b 2.746a 4.041 4.438a 3.434a Geothermal 3.871a 1.995 0.125 0.049 Hydro 11.886a 3.748 0.235 1.095b 4.763 3.453a 2.692a Solar 6.640a 0.878 1.673b 0.578 Waste 10.264a 3.240 0.745a 2.734a 4.161 6.739a 4.235a Wind 14.051a 1.349 3.855a 2.014b 2.457 6.376a 3.369a

2.310 1.149 1.035c

2.063 0.953 0.799

Tests assume an individual intercept and individual trend. Only weight statistics are reported for the within-dimension tests. Automatic lag length determined by the Schwarz Information Criteria. The null hypothesis is no cointegration. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

130

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 7 Dynamic OLS and fully modied OLS long-run estimates. DOLS Log(Renew) Log(Non-renew) Log(Capital) Log(Labor) 0.012 (1.06) 0.013c (1.33) 0.186a (11.00) 0.288a (5.11) FMOLS 0.149a (9.97)b 0.329a (10.41) 0.382a (14.60) 1.10a (12.55)

hydro, waste, and wind energy, all indicate cointergration at the 5% signicance level for the tests that require a common root across countries, and for the tests that allow countries to have an individual autoregressive process. Geothermal only rejects the null hypothesis for one test, and solar energy rejects the null hypothesis of no cointegration for two within tests. Thus, we proceed to examine the long-run equilibrium relationship for all sources, but note that geothermal and solar energy only exhibit a weak cointegrating relationship with real GDP. 2.3. Long-run equilibrium parameters To examine the parameters of the long-run relationship between renewable generation and economic growth, we estimate Eq. (1) using two methods: dynamic OLS, and fully modied OLS. DOLS works to reduce the bias from endogeneity and serial correlation found in OLS estimations. Kao and Chiang (2000) also found that DOLS outperforms OLS estimators for panel data. The fully modied OLS (FMOLS) technique proposed by Pedroni (2001) is reported for case comparison with previous studies. Estimation results for Eq. (1) are displayed in Table 7. All coefcients are positive and statistically signicant, and most notably the coefcient for renewable electricity. The DOLS results predict that a 1% increase in renewable generation increases real GDP by 0.012%. The FMOLS estimates an elasticity for renewable generation of 0.149. Apergis and Payne (2011b) report slightly higher elasticities for developed countries 0.265 and developing countries 0.429. Apergis and Payne (2011c) report an elasticity of 0.074 for emerging markets.6 The FMOLS estimates are similar to those previously found in the literature modeling the renewable energy-economic growth nexus in a production framework. Our estimated elasticity for renewable energy of 0.149 is much lower than Apergis and Payne (2012b) at 0.371, but the non-renewable estimates are comparable at 0.329 and 0.384. The estimates for capital elasticity are also similar at 0.382 and 0.388, and the labor elasticities differ at 1.10 and 0.812. Our estimates are also comparable to Apergis and Payne (2011b) who report elasticities of 0.265 for RE, 0.294 for NRE, 0.327 for capital, and 0.361 for labor, for developed countries. The long-run relationship modeled in Eq. (2) is estimated using the FMOLS technique. Table 8 presents the results for each source by column: (1) biomass, (2) geothermal, (3) hydroelectricity, (4) solar, (5) waste, and (6) wind. Noteworthy patterns emerge: a) across all renewable energy sources the estimated coefcients are positive and statistically signicant; b) a 1% increase in biomass increases real GDP by 0.129%; c) a similar increase in hydroelectricity and waste generation increases real GDP by 0.114%, and 0.096%; d) geothermal, solar, and wind have the smallest impact with estimated elasticities of 0.085, 0.055, and 0.053; and e) the elasticities for non-biomass (0.41), nongeothermal (0.53), non-hydro (0.38), non-solar (0.48), and non-waste (0.53) variables are comparable to the non-renewable measure reported in Table 7 of 0.33. Interestingly, non-wind generation has an elasticity of 3.86. In the next sections, we proceed to test for causality between each source and real GDP. 3. Results for the panel error correction models Using a panel ECM, we examine the causal relationship between renewable electricity and economic activity. The rst difference of each variable is modeled as a function of the lagged difference of each explanatory variable and an error correction term (ECT) based on the
6 Apergis and Payne (2010b) report an elasticity of 0.20 for Eurasian countries with Russia and 0.07 without. Apergis and Payne (2010) reports 0.76 for OECD countries, and Apergis and Payne (2011d) estimates 0.24 for Central American countries. These papers exclude NRE as a control.

Long-run estimates from Eqs. (3a)(3e) are provided in this table. t-statistic reported in the parenthesis. N = 380. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

lagged residuals from Eq. (1). One lag-length is utilized, such that the model becomes Yit 1i 11i Yit1 12i REit1 13i NREit1 14i Kit1 15i Lit1 1i 1it1 u1it 3a

REit 2i 21i Yit1 22i REit1 23i NREit1 24i Kit1 25i Lit1 2i 2it1 u2it

3b

NREit 3i 31i Yit1 32i REit1 23i NREit1 34i Kit1 35i Lit1 3i 3it1 u3it

3c

Kit 4i 41i Yit1 42i REit1 43i NREit1 44i Kit1 45i Lit1 4i 4it1 u4it 3d

Lit 5i 51i Yit1 52i REit1 53i NREit1 54i Kit1 55i Lit1 5i 5it1 u5it 3e

where the ECTs are modeled as the lagged residuals from Eq. (1). A negative and statistically signicant coefcient for the ECT, , implies the variables are cointegrated and shocks to the model generate movements back toward equilibrium. A positive (negative) and statistically signicant result for each kj coefcient indicates that the variable has a positive (negative) short-run causal impact on the dependent variable. Our panel ECM is estimated using the method developed by Pesaran et al. (1999), which accounts for hetergoneous panels. We test for efciency using a Hausman test, which indicates the meangroup estimator is preferred over the pooled mean group and dynamic xed-effects estimators. The mean-group estimator allows the long-run estimates to vary by country, and the results presented are of the unweighted means of the coefcients. The results for the panel ECM for renewable generation are presented in Table 9. Each row represents an equation from the model with the columns capturing the different sources of causation in the short-run by the independent variables and the long-run by the ECT. We provide test statistics for a test of inclusion for each variable. In parenthesis, we report the estimated coefcient for each variable and below in brackets, we report the p-value from a likelihood ratio test with a null hypothesis of H0: kj = 0. All variables have a short-run Granger-causing impact on the dependent variable. For real GDP in row 1, the results imply that renewable electricity, non-renewable electricity, capital, and labor are signicant sources of causation in the short-run. For renewable generation in

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 8 Fully modied OLS for each renewable sources. Log(GDP) (1) Log(Bio) Log(Non-bio) Log(Geo) Log(Non-geo) Log(Hydro) Log(Non-hydro) Log(Solar) Log(Non-solar) Log(Waste) Log(Non-waste) Log(Wind) Log(Non-wind) Log(Capital) Log(Labor) N 0.129a (15.21) 0.412a (10.77) 0.325a (17.15) 0.903a (11.39) 342 Log(GDP) (2) 0.085a (6.43) 0.533a (7.48) 0.251a (8.15) 1.184a (8.73) 114 Log(GDP) (3) 0.114a (5.96) 0.384a (11.63) 0.378a (17.99) 1.220a (11.07) 380 Log(GDP) (4) 0.055a (15.20) 0.478a (3.90) 0.294a (9.25) 0.168b ( 2.04) 114 Log(GDP) (5) 0.096a (12.19) 0.528a (10.80) 0.296a (10.94) 1.156a (13.07) 304

131

Log(GDP) (6) 0.053a (23.13) 3.864a (6.64) 0.321a (11.73) 0.822a (9.91)c 190

Estimates for capital and labor are excluded for brevity. t-statistic reported in the parenthesis. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

row 2, the results suggest that GDP, non-renewable electricity, capital, and labor are signicant sources. These results provide evidence of bidirectional causality between renewable generation and real GDP, supporting the feedback hypothesis. Given the nature of our estimation, the ECT coefcient should be negative for all equations, indicating that deviations from equilibrium generate movements back toward the long-run equilibrium. In Table 9, the ECT is statistically signicant for all equations suggesting that economic growth, renewable and nonrenewable generation, capital,

and labor are cointegrated. The speed of adjustment for GDP is 10.6 months, and 8.4 months for renewable energy, which is much quicker than that suggested by Apergis and Payne (2011b, 2012). Thus, the data provide evidence of both a short- and long-run relationship between renewable generation and economic growth, supporting much of the literature that includes both renewable and nonrenewable energy, capital, and labor. Next, we examine the dynamics of GDP growth with individual sources using the corresponding panel ECM. In a manner similar to the

Table 9 Panel causality tests for aggregate renewable measure. Sources of Causation (independent variables) Short-run Dependent variable (3a) GDP (3b) RE (3c) NRE (3d) K (3e) L GDP 56.80 (0.848) [0.00]a 71.63 (0.209) [0.00]a 167.34 ( 0.075) [0.00]a 111.23 ( 0.117) [0.00]a RE 89.74 (0.0385) [0.00]a 44.86 ( 0.228) [0.00]a 70.09 (0.050)b [0.00]a 65.38 (0.020) [0.00]a NRE 64.56 ( 0.013) [0.00]a 71.22 (0.119) [0.00]a 67.43 ( 0.096)c [0.00]a 75.03 ( 0.028) [0.00]a Long-run K 106.42 (0.007) [0.00]a 55.45 (0.066) [0.00]a 52.08 (0.298) [0.00]a 130.79 (0.054) [0.00]a L 102.29 ( 0.141) [0.00]a 88.36 (2.045) [0.00]a 86.68 (0.744) [0.00]a 121.74 (0.183) [0.00]a ECT 1.129 [0.000]a 1.429 [0.000]a 1.053 [0.000]a 0.487 [0.026]a 0.665 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

132 Table 10a Panel causality test for biomass. GDP (4a) GDP (4b) Bio (4c) Non-biomass (4d) K (4e) L 102.77 (1.901) [0.00]a 78.76 ( 0.396) [0.00]a 149.79 ( 0.374) [0.00]a 148.91 (0.074)c [0.00]a

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139

Bio 112.23 ( 0.014) [0.00]a 102.75 ( 0.074) [0.00]a 95.75 ( 0.082) [0.00]a 152.56 (0.003) [0.00]a

Non-biomass 109.38 (0.145) [0.00]a 90.78 (0.695) [0.00]a 107.97 (0.408) [0.00]a 71.11 (0.026) [0.00]a

K 141.84 (0.031) [0.00]a 83.21 (0.015) [0.00]a 103.49 (0.080) [0.00]a 127.99 (0.022) [0.00]a

L 155.03 (0.145) [0.00]a 70.50 (3.760) [0.00]a 69.79 (0.486) [0.00]a 105.01 (0.894) [0.00]a

EC 1.032 [0.000]a 1.053 [0.000]a 1.369 [0.000]a 0.460 [0.014]b 0.603 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

Table 10b Panel causality test for geothermal. GDP (5a) GDP (5b) Geo (5c) Non-geothermal (5d) K (5e) L 26.41 ( 3.124) [0.006]b 30.61 ( 0.339) [0.00]b 63.56 ( 1.005) [0.00]b 59.12 ( 0.056) [0.00]b Geothermal 23.69 ( 0.074) [0.01]a 7.81 ( 0.036) [0.73]c 28.64 ( 0.234) [0.00]b 33.08 ( 0.027) [0.00]b Non-geothermal 15.77 ( 0.030) [0.15] 10.42 ( 1.288) [0.49] 11.36 (0.178) [0.41] 4.22 ( 0.083) [0.96] K 79.09 (0.110) [0.00]b 20.61 (1.416) [0.04]a 42.49 (0.238) [0.00]b 51.73 ( 0.010) [0.00]b L 40.84 ( 0.059) [0.00]b 20.67 (5.503) [0.04]a 20.34 (0.223) [0.04]a 43.86 (0.726) [0.00]b EC 1.562 [0.000]b 1.270 [0.000]b 1.230 [0.000]b 0.387 [0.204] 0.696 [0.000]b

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .05 level. b Statistical signicance at .01 level. c Statistical signicance at .10 level.

aggregate renewable energy specication, the model for biomass becomes Yit 1i 11i Yit1 12i BIOit1 13i NonBIOit1 14i Kit1 15i Lit1 1i 1it1 u1it BIOit 2i 21i Yit1 22i BIOit1 23i NonBIOit1 24i Kit1 25i Lit1 2i 2it1 u2it NonBIOit 3i 31i Yit1 32i BIOit1 23i NonBIOit1 34i Kit1 35i Lit1 3i 3it1 u3it Kit 4i 41i Yit1 42i BIOit1 43i NonBIOit1 44i Kit1 45i Lit1 4i 4it1 u4it Lit 5i 51i Yit1 52i BIOit1 53i NonBIOit1 54i Kit1 55i Lit1 5i 5it1 u5it: 4a

4b

4c

4d

4e

where the ECT is the lagged residuals from Eq. (2). For the other ve sources, we use a similar model, and notate the equations as 5a5e for geothermal, 6a6e for hydroelectric, 7a7e for solar, 8a8e for waste, and 9a9e for wind. Table 10a10f reports the panel causality results by source, where each panel (af) represents a new source. The columns display the short and long-run sources of causation for each equation. Examining the partial-F statistics in row 1 for each panel, we note the short-run

impact of each energy source on GDP. Biomass, geothermal, hydroelectricity, waste and wind exhibit a statistically signicant short-run causal relationship with GDP. In Table 10a, the estimated coefcient of 0.014 from the GDP estimation (4a) indicates that short-run increases in biomass decrease the growth of real GDP. Eq. (4b) for biomass reports that real GDP has a positive short-run impact on biomass. Thus, a bidirectional relationship exists between biomass and GDP.7 For hydroelectricity and waste energy in Tables 10c and 10e, the estimated coefcients for each renewable source in Eqs. (6a) and (8a) indicate a positive short-run impact on real GDP. Eqs. (6b) and (8b) report a positive short-run causal relationship from real GDP to the renewable energy source. These results indicate a positive bidirectional relationship, supporting the feedback hypothesis between hydroelectricity and GDP, and between waste energy and GDP. In Tables 10b and 10f, the results for geothermal and wind report short-run causality, but both exhibit negative bidirectional causality with real GDP, as noted by Eqs. (5a) and (5b) for geothermal, and Eqs. (9a) and (9b) for wind energy. Finally, solar energy in Table 10d exhibits only unidirectional causality from GDP to solar electricity generation. Thus, hydroelectricity and waste energy are two important factors in the renewable energy GDP nexus.

7 This result opposes the ndings of Payne (2011), which found evidence of a unidirectional relationship, where increases in biomass consumption positively affect real GDP for the United States.

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 10c Panel causality test for hydro. GDP (6a) GDP (6b) Hydro (6c) Non-hydro (6d) K (6e) L 34.95 (0.524) [0.00]a 57.32 (0.282) [0.00]a 147.95 ( 0.207) [0.00]a 119.56 ( 0.136) [0.00]a Hydro 68.84 (0.062) [0.00]a 66.15 ( 0.147) [0.00]a 57.27 (0.148) [0.00]a 77.51 (0.007) [0.00]a Non-hydro 65.32 ( 0.001) [0.00]a 64.01 (0.224) [0.00]a 74.83 ( 0.083) [0.00]a 65.09 ( 0.030) [0.00]a K 98.08 ( 0.009) [0.00]a 55.09 (0.038) [0.00]a 64.29 (0.260) [0.00]a 140.62 (0.018)c [0.00]a L 104.93 (0.057) [0.00]a 81.02 (0.590) [0.00]a 71.83 (0.350) [0.00]a 105.29 (0.229) [0.00]a EC

133

1.154 [0.000]a 1.488 [0.000]a 1.142 [0.000]a 0.47 [0.022]b 0.613 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

Table 10d Panel causality test for solar. GDP (7a) GDP (7b) Solar (7c) Non-solar (7d) K (7e) L 27.55 (18.783) [0.00]a 34.21 (0.574) [0.00]a 34.5 ( 0.058) [0.00]a 7.20 ( 0.186) [0.78] Solar 8.79 (0.007) [0.64] 42.42 ( 0.138) [0.00]a 24.10 ( 0.033) [0.01]b 3.00 (0.002) [0.99] Non-solar 26.61 ( 0.073) [0.005]a 3.02 (1.922) [0.99] 32.74 ( 0.435) [0.00]a 16.70 ( 0.011) [0.12] K 22.25 (0.030) [0.02]b 24.07 ( 4.545) [0.01]b 9.68 ( 0.365) [0.56] 34.34 (0.068) [0.00]a L 37.13 (0.028) [0.00]a 23.35 (5.850) [0.02]b 35.52 (0.318) [0.00]a 28.02 ( 0.557) [0.00]a EC 1.222 [0.000]a 1.391 [0.000]a 1.240 [0.000]a 0.237 [0.173] 0.477 [0.056]c

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

Analysis of the ECT, , also supports the idea that biomass, hydroelectricity, and waste energy are driving factors in the renewable energyGDP relationship. The estimated coefcient for is negative and statistically signicant across all equations (ae) for several energy sources, including biomass (Eqs. (4a)(4e)), hydroelectricity (Eqs. (6a)(6e)), and waste energy (Eqs. (8a)(8e)), consistent with a long-run cointegrating relationship between real GDP, capital, labor, and the energy source. Geothermal, solar, and wind all have a positive and/or insignicant ECT for capital in Eqs. (5d), (7d) and (9d). As noted before, geothermal and solar may not exhibit a cointegrating relationship with GDP. These results conrm the ndings of a bidirectional relationship and provide support for the feedback hypothesis between renewable energy
Table 10e Panel causality test for waste. GDP (8a) GDP (8b) Waste (8c) Non-waste (8d) K (8e) L 52.75 (2.706) [0.00]a 57.44 ( 0.896) [0.00]a 158.06 (0.752) [0.00]a 92.44 ( 0.012) [0.00]a Waste 42.64 (0.029) [0.00]a 67.05 (0.021) [0.00]a 87.14 (0.054) [0.00]a 68.47 ( 0.0001) [0.00]a

and real GDP. All sources do exhibit a long-run relationship with GDP, but only hydroelectricity and waste generation exhibit a strong positive bidirectional relationship that supports the feedback hypothesis. Furthermore, biomass, hydroelectricity, and waste energy exhibit the largest long-run impact on real GDP. 4. Additional analysis for structural breaks and cross-sectional dependence Extending the previous renewable energyeconomic growth literature, we test for unit roots with structural breaks, and we examine the model accounting for cross-sectional dependence (CSD). Structural

Non-waste 70.22 (0.061) [0.00]a 54.42 (0.504)b [0.00]a 69.06 (0.013) [0.00]a 48.08 (0.004) [0.00]a

K 118.19 ( 0.091) [0.00]a 63.70 (0.042) [0.00]a 85.19 (0.196) [0.00]a 68.47 (0.009) [0.00]a

L 69.62 (0.146) [0.00]a 89.38 (1.327) [0.00]a 36.30 ( 0.331) [0.00]a 93.85 ( 0.384) [0.00]a

EC 1.438 [0.000]a 0.887 [0.000]a 1.360 [0.000]a 0.414 [0.070]c 0.632 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

134 Table 10f Panel causality test for wind. GDP (9a) GDP (9b) Wind (9c) Non-wind (9d) K (9e) L 50.15 ( 4.487) [0.00]a 46.35 ( 0.078) [0.00]a 94.15 (0.828) [0.00]a 45.75 ( 0.147) [0.00]a

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139

Wind 43.80 ( 0.003) [0.00]a 29.90 (0.001)b [0.00]a 39.46 ( 0.023) [0.00]a 30.71 (0.004) [0.00]a

Non-wind 43.96 ( 0.847) [0.00]a 34.41 (5.576) [0.00]a 51.67 ( 5.736) [0.00]a 18.57 ( 0.513) [0.07]c

K 66.37 ( 0.002) [0.00]a 40.85 (1.058) [0.00]a 39.62 (0.030) [0.00]a 34.34 (0.073) [0.00]a

L 70.25 ( 0.027) [0.00]a 61.46 ( 1.565) [0.00]a 57.52 (0.024) [0.00]a 88.34 ( 0.547) [0.00]a

EC 1.464 [0.000]a 0.577 [0.003]a 1.357 [0.000]a 0.358 [0.101] 0.519 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. p-Values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

breaks occur when shocks to the series cause a permanent change. CSD occurs when renewable electricity generation is not independent between countries. This may likely be the case for countries where the electricity grid is interconnected, and policies in one country may affect the electricity generation in another country. Thus, we examine the panel ECM controlling for CSD, but due to the limited amount of data, we forgo the production model framework and exclude capital and labor from the analysis.

non-renewable, non-biomass, hydro, non-hydro, solar, non-solar, nonwaste, and non-wind are integrated of order one.

Table 11 Structural breaks for aggregate renewable energy. AO 1 break Australia 2000 (0.06) 1994 (0.00) 2001 (0.00) 2005 (0.02) 1999 (0.00) 1999a (0.59) 2001 (0.00) 2001 (0.00) 1997a (0.00) 1992 (0.93) 1999 (0.00) 2001 (0.02) 1999a (0.19) 2006a (0.02) 1994a (0.02) 1997 (0.00) 1994a (0.66) 1999a (0.35) 2001 (0.00) 1999 (0.34) IO 1 break 2004 (0.07) 1996 (0.11) 2003 (0.07) 2003 (0.05) 1995a (0.00) 2000 (0.13) 1998 (0.07) 2005 (0.01) 1997a (0.00) 1996a (0.07) 1996 (0.04) 1994 (0.39) 1992a (1.00) 1998a (0.01) 1994a (0.00) 1999 (0.02) 1995a (0.26) 2000 (0.68) 1997 (0.23) 2000 (0.40) AO 2 breaks 1996, 2003 (0.00), (0.04) 1998, 2000 (0.01), (0.26) 1998, 2004 (0.00), (0.00) 1993, 2004a (0.00), (0.00) 1994, 1999 (0.00), (0.00) 1999, 2003 (0.88), (0.37) 1999, 2004 (0.00), (0.00) 1998, 2003 (0.00), (0.01) 1997, 1999a (0.15), (0.56) 1991, 1996a (0.10), (0.02) 1998, 2004 (0.00), (0.00) 1997, 2003 (0.00), (0.00) 1998, 2001 (0.70), (0.11) 1998, 2006a (0.09), (0.06) 1994, 2001a (0.05), (0.83) 1994, 2001 (0.02), (0.01) 1994, 1998 (0.66), (0.22) 1999, 2003 (0.11), (0.16) 1998, 2003 (0.00), (0.00) 1992, 1999 (0.33), (0.21)
a

IO 2 breaks 1995, 2004a (0.01), (0.01) 1997, 2002 (0.06), (0.77) 1996, 2003 (0.01), (0.59) 1994, 2003 (0.03), (0.01) 1995, 2001 (0.00), (0.06) 1993, 2000 (0.62), (0.20) 1998, 2005 (0.04), (0.23) 1995, 2005 (0.21), (0.01) 1998, 2002a (0.01), (0.88) 1992, 1996a (0.04), (0.00) 1998, 2004 (0.01), (0.01) 1994, 2002 (1.00), (0.02) 1992, 2000 (1.00), (0.01) 1996, 2003 (0.10), (0.06) 1994, 1997a (0.00), (0.99) 1994, 2001 (1.00), (0.01) 1995, 2006 (0.38), (0.51) 1997, 2003a (0.01), (0.02) 1996, 2003 (0.00), (0.01) 1993, 2000 (0.16), (0.89)

4.1. Structural breaks Using the time series tests developed by Perron and Vogelsang (1992) and Clemente et al. (1998), we test each country's renewable generation for a unit root with one and two structural breaks. The additive outlier (AO) and innovational outlier (IO) are two approaches to modeling the structural break, where the AO test models the break as an abrupt change, such as a policy change, and the IO test models the break as a gradual change, such as a change in technology with slow adoption. Table 11 reports the optimal structural break for each country. In parenthesis is the p-value for that break's estimated coefcient, and the superscript notates the unit root test result. For most countries, we fail to reject the presence of a unit root, and most structural breaks are statistically signicant, occurring between 1997 and 2002. For several countries not in the EU (Canada, Japan, New Zealand, United States), the structural break falls outside these dates and/or is not statistically signicant. We note that in 1997 the EU signed the Amsterdam Treaty, which promoted sustainable development and encouraged renewable generation. Thus, it is likely that the structural break is caused by a common factor, and we consider the possibility of CSD. Given that several countries have structural breaks, we test for panel stationarity allowing for multiple unknown breaks. Table 12 reports the results of a test proposed by Carrion-i-Silvestre et al. (2005), which provides exibility in the number of breaks, as well as allowing the break points to differ across countries. Our results indicate the presence of a unit root, even when allowing for structural breaks. Thus, the variables of interest remain non-stationary at levels.

Austria Belgium Canada Denmark France Germany Iceland Italy Japan Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States

4.2. Cross-sectional dependence We test for CSD using the method developed in Pesaran (2004). Each of the energy sources are tested for CSD and the average correlation across countries is reported in Table 13. The results indicate CSD for all renewable energy sources, non-renewable energy, and GDP. Next, we implement the Pesaran (2007) a unit root test that controls for CSD. Table 14 reports the results that GDP, capital, and labor are trend stationary when accounting for CSD. Only renewable,

This table reports the optimal break dates for the series. In parenthesis is the p-value for the estimated coefcient of each break. The additive outlier (AO) model assumes changes take place rapidly. The innovational outlier (IO) model assumes changes take place gradually. a Indicates the null hypothesis of a unit root is rejected because the t-statistic is less than the critical value: 3.560 for AO with 1 break, 4.270 for IO with 1 break, and 5.49 for AO and IO with 2 breaks. The test is with a 5% signicance level.

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 12 Panel unit root tests with structural breaks. Bartlett kernel Homogeneous breaks Log(GDP) Log(Renewable) Log(Non-renewable) Log(Biomass) Log(Geothermal) Log(Hydro-electric) Log(Solar) Log(Waste) Log(Wind) 42.813 (0.000) 18.107 (0.000) 28.811 (0.000) 17.133 (0.000) 7.206 (0.000) 12.686 (0.000) 41.207 (0.000) 32.102 (0.000) 42.004 (0.000) Heterogeneous breaks 264.153 (0.000) 104.895 (0.000) 156.125 (0.000) 301.957 (0.000) 183.966 (0.000) 49.173 (0.000) 60.642 (0.000) 282.286 (0.000) 122.868 (0.000) Quadratic kernel Homogeneous breaks 43.476 (0.000) 18.279 (0.000) 29.201 (0.000) 16.814 (0.000) 7.902 (0.000) 12.596 (0.000) 45.126 (0.000) 31.095 (0.000) 41.839 (0.000)

135

Heterogeneous breaks 267.844 (0.000) 99.315 (0.000) 176.583 (0.000) 288.526 (0.000) 181.291 (0.000) 46.047 (0.000) 61.642 (0.000) 271.380 (0.000) 119.541 (0.000)

p-Values are reported in parenthesis. The tests allow up to 5 structural breaks. The null hypothesis is panel stationarity.

Table 13 Cross-sectional dependence test statistic and average correlation. Pesaran CSD test Log(GDP) Log(Renewable) Log(Non-renewable) Log(Biomass) Log(Geothermal) Log(Hydro-electric) Log(Solar) Log(Waste) Log(Wind) 56.88 (0.00) 24.85 (0.00) 31.80 (0.00) 40.62 (0.00) 6.60 (0.00) 7.08 (0.00) 8.75 (0.00) 39.19 (0.00) 26.97 (0.00) Average correlation 0.947 0.414 0.529 0.753 0.391 0.118 0.518 0.821 0.922

REit 2i 21i Yit1 22i REit1 23i NREit1 v21i Yit v22i REit v23i NREit 2i 2it1 u2it
CSD

10b

NREit 3i 31i Yit1 32i REit1 33i NREit1 v31i Yit v32i REit v33i NREit 3i 3it1 u1it
CSD

10c where the ECT is modeled as the lagged residuals of the long-run equation that accounts for CSD Yit 1i 2i t 3i REit 4i NREit 1i Y 2i RE 3i NRE :

p-Values are reported in parenthesis. The null hypothesis is cross-sectional independence.

Westerlund and Edgerton (2007) propose four cointegration tests, and control for CSD using a bootstrapping method for the report p-values. Table 15 reports the results that each energy source does not exhibit a cointegrating relationship between GDP, the renewable source, and the source's complement. Thus, using this approach to testing for cointegration, each of the renewable sources do not appear to be cointegrated with GDP. Thus, we further test for cointegration using the panel ECM while controlling for CSD. 4.3. Mean group panel error correction model corrected for cross-sectional dependence The mean group panel ECM is used to further test for cointegration, and examine possible short run causality. Following the model proposed by Pesaran (2006) and Binder and Offermanns (2007), we account for unobserved common factors by including the cross-country average of each variable.8 The model becomes
Yit 1i 11i Yit1 12i REit1 13i NREit1 v11i Y v12i REit v13i NREit 1i 1it1 u1it
CSD

it

CSD

10a
8 Due to the limited number of observation and estimated parameters, capital and labor are excluded.

For each individual source, we use a similar model, and notate the equations as 11a11c for biomass, 12a12c for geothermal, 13a13c for hydroelectric, 14a14c for solar, 15a15c for waste, and 16a16c for wind. Tables 16a and 16g present the results from the panel ECM. Examining the partial-F statistics in row 1 for each panel, we note the short-run impact of each energy source on GDP. All sources exhibit a statistically signicant short-run causal relationship with GDP, except solar energy. In Table 16a, aggregate renewable energy exhibits a bidirectional relationship as before, but the estimated coefcients indicate a change in sign. Increases in renewable energy increase GDP, but increases in GDP decrease renewable generation. The results for biomass, geothermal, and solar energy remain unchanged. The estimated coefcient of 0.018 in Table 16b Eq. (11a) indicates that short-run increases in biomass decrease the growth of real GDP. Eq. (4b) for biomass reports that real GDP has a positive short-run impact on biomass. Table 16c reports geothermal exhibits a short-run negative bidirectional causal relationship with GDP. Finally, solar energy in Table 16e exhibits only unidirectional causality from GDP to solar electricity generation. For hydroelectricity and waste energy in Tables 16d and 16f, we still nd bidirectional causality, but the estimated coefcients for each source changes. Increases in hydroelectricity increase GDP, but increases in GDP decrease hydroelectricity. The reverse is the case for

136 Table 14 Unit root test with cross-sectional dependence and trend. GDP Lag(0) Lag(1) 0.660 (0.745) 2.34 (0.990) GDP Lag(0) Lag(1) 0.098 (0.539) 2.544 (0.995) GDP Lag(0) Lag(1) 1.281 (0.100) 1.374 (0.915) GDP Lag(0) Lag(1) 0.660 (0.745) 2.34 (0.990) GDP Lag(0) Lag(1) 2.183 (0.985) 1.613 (0.947) GDP Lag(0) Lag(1) 0.217 (0.586) 1.809 (0.965) GDP Lag(0) Lag(1) 1.576 (0.057) 1.171 (0.121) Renewable 4.375 (0.000) 0.721 (0.235) Biomass 0.089 (0.535) 0.926 (0.823) Geothermal 0.589 (0.278) 0.008 (0.503) Hydro 3.519 (0.000) 0.631 (0.264) Solar 2.284 (0.011) 0.256 (0.399) Waste 1.040 (0.851) 1.068 (0.857) Wind 1.678 (0.953) 1.042 (0.851)

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 16a Panel causality test for renewable without capital and labor. Non-renew 3.225 (0.001) 0.641 (0.739) Non-bio 4.128 (0.000) 1.321 (0.093) Non-geo 1.829 (0.966) 2.659 (0.996) Non-hydro 2.435 (0.007) 0.327 (0.628) Non-solar 1.529 (0.063) 0.748 (0.227) Non-waste 2.471 (0.007) 1.224 (0.890) Non-wind 2.828 (0.002) 2.006 (0.022) (10a) GDP (10b) Renew (10c) Non-renew GDP 104.18 ( 2.233) [0.00]a 166.35 ( 0.144) [0.00]a Renew 185.65 (0.042) [0.00]a 121.66 ( 0.218) [0.00]a Non-renew 111.79 (0.085) [0.00]a 97.92 (0.367) [0.00]a EC 0.612 [0.000]a 1.313 [0.000]a 0.813 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level.
b c

Statistical signicance at .05 level. Statistical signicance at .10 level.

Table 16b Panel causality test for biomass without capital and labor. GDP (11a) GDP (11b) Bio (11c) Non-biomass 168.27 (0.401) [0.00]a 127.03 (0.151) [0.00]a Bio 134.45 ( 0.018) [0.00]a 167.98 (0.070) [0.00]a Non-biomass 119.35 ( 0.016) [0.00]a 181.54 (0.730) [0.00]a EC 0.895 [0.000]a 1.486 [0.000]a 1.288 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level.
b c

Statistical signicance at .05 level. Statistical signicance at .10 level.

These results highlight the complicated nature of the relationship of renewable energy and GDP, and the importance of disaggregating sources to examine the individual impacts. Future research may examine the impact of CSD in a production model framework. 5. Conclusions This study contributes to previous research on the causal relationship between renewable energy and economic growth by examining electricity generation by each renewable source: biomass, geothermal, hydroelectricity, solar, waste, and wind energy. Using data from 20 OECD countries from 1990 to 2008, we implement several panel cointegration tests. Similar to previous studies, we report a long-run relationship between real GDP, renewable generation, non-renewable energy, real
Table 16c Panel causality test for geothermal without capital and labor. GDP (12a) GDP (12b) Geo (12c) Non-geothermal 41.79 ( 1.618) [0.00]a 51.18 (0.230) [0.00]a Geothermal 64.79 ( 0.046) [0.00]a 19.22 ( 0.021) [0.08]c Nongeothermal 34.80 (0.205) [0.00]a 36.37 (0.959) [0.00]a EC 0.229 [0.092]c 0.869 [0.005]a 1.304 [0.001]a

This table reports the Z t-bar statistics for the Pesaran (2007) unit root test controlling for cross-sectional dependence, including a trend. p-Values are reported in parenthesis. All variables are log-transformed.

Table 15 Westerlund error correction model panel cointegration test, robust p-values. Aggregate renew Gt Ga Pt Pa 0.724 0.776 0.698 0.710 Biomass 0.812 0.408 0.564 0.606 Geothermal 0.670 0.802 0.620 0.610 Hydro 0.872 0.968 0.742 0.758 Solar 0.628 0.632 0.506 0.502 Waste 0.908 0.754 0.838 0.830 Wind 0.418 0.242 0.212 0.138

Bootstrapped p-values at 500.

waste generation. Increases in waste decrease GDP, but increases in GDP cause increases in waste.9 Analysis of the ECT, , supports the idea of a cointegrating relationship between renewable energy and GDP when accounting for CSD. The estimated coefcient for is negative and statistically signicant across all equations (ac) for all energy sources in Tables 16a16g, consistent with a long-run cointegrating relationship between real GDP, the energy source, and all other sources. The speed of adjustment for GDP is 13.4 months for biomass, 52 for geothermal, 20 for hydro, 24 for solar, 15 for waste, and 14 for wind, which is much slower than the results from Section 3, but perhaps somewhat more realistic.
9

The appendix includes the time and xed effects coefcients for this model.

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table 16d Panel causality test for hydro without capital and labor. GDP (13a) GDP (13b) Hydro (13c) Non-hydro 115.26 ( 1.878) [0.00]a 108 ( 0.023) [0.00]a Hydro 177.79 (0.064) [0.00]a 100.08 ( 0.051) [0.00]a Non-hydro 152.03 ( 0.033) [0.00]a 176.56 ( 0.522) [0.00]a EC 0.611 [0.000]a 1.613 [0.000]a 0.752 [0.007]a (16a) GDP (16b) Wind (16c) Non-wind Table 16g Panel causality test for wind without capital and labor. GDP 77.06 (3.390) [0.00]a 91.83 (0.013) [0.00]a Wind 70.36 (0.010) [0.00]a 178.77 ( 0.007) [0.00]a Non-wind 51.34 (1.113) [0.00]a 85.33 (6.729)b [0.00]a EC

137

0.855 [0.005]a 0.847c [0.007]a 1.641 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level.
b c

Statistical signicance at .05 level. Statistical signicance at .10 level.

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

Table 16e Panel causality test for solar without capital and labor. GDP (14a) GDP (14b) Solar (14c) Non-solar 117.79 (7.595) [0.00]a 17.10 ( 0.025) [0.15] Solar 15.21 (0.002) [0.23] 38.50 ( 0.022) [0.00]a Non-solar 35.12 (0.137) [0.00]a 49.14 ( 1.686) [0.00]a EC 0.507 [0.031]b 1.275 [0.003]a 2.149 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level. b Statistical signicance at .05 level. c Statistical signicance at .10 level.

gross xed capital, and the labor force. The long-run elasticities from a fully modied OLS model are positive and statistically signicant. Using a panel ECM to test for causality, we nd short-run bidirectional causality between renewable energy and real GDP, supporting the feedback hypotheses. For the individual sources of renewable energy, we utilize the same cointegration tests and FMOLS estimation method. All sources exhibit a long-run relationship with real GDP, capital, and labor; but, we nd only weak evidence of a long-run cointegrating relationship between geothermal energy and GDP, and solar energy and GDP. The elasticities estimated indicate that biomass, hydroelectricity, and waste energy have the largest impact on real GDP with elasticities of 0.129, 0.114, and 0.096. Geothermal, solar, and wind exhibit the smallest impact with estimated elasticities of 0.085, 0.055 and 0.053. Utilizing the panel ECM for each individual source, we nd that hydroelectricity and waste energy are the only two sources to exhibit
Table 16f Panel causality test for waste without capital and labor. GDP (15a) GDP (15b) Waste (15c) Non-waste 62.30 (2.193) [0.00]a 146.16 ( 0.152) [0.00]a Waste 80.0 ( 0.015) [0.00]a 159.67 ( 0.158) [0.00]a Non-waste 162.36 (0.080) [0.00]a 78.19 ( 1.343) [0.00]a EC 0.776 [0.000]a 0.576 [0.004]a 1.336 [0.000]a

This table reports the partial F-statistics with respect to short-run changes in the independent variables from a likelihood ratio test. The lag length is one based on the Schwarz information criterion, and the coefcients are reported in parenthesis. P-values are reported in brackets below for the likelihood ratio test. For the ECT, we report the Wald test to determine long-run causality. a Statistical signicance at .01 level.
b c

Statistical signicance at .05 level. Statistical signicance at .10 level.

short-run bidirectional causality with real GDP, which supports the feedback hypothesis found in the aggregate measure of renewable energy. The results for biomass report short-run causality, but we report a negative causal effect from biomass to GDP and a positive impact from GDP to biomass. Geothermal and wind both exhibit negative bidirectional causality. Finally, solar energy exhibits only unidirectional causality from GDP to solar electricity generation. We extend our analysis with the panel ECM to control for crosssectional dependences, but without the production model framework. The results for biomass, geothermal, and solar energy remain unchanged. For hydroelectricity and waste energy, we still nd bidirectional causality, but the estimated coefcients for each renewable energy source changes. Increases in hydroelectric generation increase GDP, but increases in GDP decrease hydroelectric generation. The reverse is the case for waste generation. The growing attention over renewable energy is due to concern over environmental quality standards, greenhouse gas emissions, and global climate change; yet, the contribution of renewable energy to environmental quality is questionable. Salim and Raq (2012) report bidirectional causality between pollution and renewable energy, and Apergis et al. (2010) nd that renewable energy consumption has a positive impact on CO2 emissions, suggesting that renewable energy growth does not contribute to reductions in emissions. They attribute these ndings to inadequate storage and intermittency issues of renewables, suggesting that renewable sources increase the need for backup power from fossil fuels (Marques and Fuinhas, 2012). The results presented in our paper suggest that increases in renewable energy may not necessarily lead to decreases in CO2, because biomass and waste generation are important components in the renewable energyGDP nexus. Biomass and waste generation emit CO2, and by only examining the aggregate measure of renewable energy, researchers can reach misleading interpretations that renewable energy sources are not environmentally friendly and do not reduce CO2 emissions. Such an interpretation does not consider that sustainable biomass and waste energy are net neutral, releasing carbon that had been previously extracted from the atmosphere rather than the ground. Additionally, the distinction between carbon-neutral and non-carbon neutral biomass and waste energy production may provide further insight into renewables' impact on CO2 emissions. The results of this paper are important because they illustrate that individual sources of renewable energy differ in impact on economic growth, as well as their impact on the environment. When developing environmentally friendly or economically benecially energy policies, it is important that policymakers take both into consideration. Energy conservation policies may negatively impact GDP, if the policies cause decreases in hydroelectricity energy, but not necessarily for biomass, geothermal, solar, or waste energy. Future research should examine the interaction between the sources of energy, as well as their relationship with GDP.

138

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139

Appendix A
Table A1 Time effects from the ECM accounting for cross-sectional dependence. Aggregate renewable Australia Austria Belgium Canada Denmark France Germany Iceland Italy Japan Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States Group mean
a b c

Biomass 0.117 (0.141) 1.309 (1.573) 0.203 (0.206) 0.131a (0.015) 0.568b (0.283) 0.0971 (0.130) 0.0880 (0.057) 0.0802 (0.208) 0.288a (0.060) 0.167 (0.685) 0.215a (0.041) 0.0108 (0.163) 0.0278b (0.013) 0.377 (0.517) 0.0707 (0.087) 0.109 (0.393) 0.100 (0.469) 0.0185 (0.036) 0.0156 (0.090)

Geothermal 0.0486 (0.217) 0.00301 (0.007) 0.000533 (0.034) 0.0347b (0.017) 0.292 (0.178) 0.0119 (0.041) 0.0373 (0.052)

Hydro 0.0120 (0.013) 0.0171 (0.101) 0.0102 (0.034) 0.00870 (0.010) 0.0108 (0.212) 0.0287 (0.071) 0.929 (7.363) 0.0368 (0.091) 3.610 (17.010) 0.210 (0.609) 0.113 (0.720) 0.648c (0.361) 0.0400a (0.009) 0.110a (0.020) 0.128 (0.347) 0.161 (0.216) 0.172b (0.067) 0.131 (0.235) 0.186b (0.082) 0.152 (0.132) 0.163 (0.192)

Solar 0.0744 (0.235) 0.0989 (0.329) 0.123a (0.039) 0.333a (0.081) 0.0960 (0.657) 6.127 (53.244) 1.076 (1.012)

Waste 2.813 (2.654) 0.183 (0.166) 0.0720b (0.037) 1.742 (3.986) 1.068 (1.890) 0.131 (0.248) 0.112 (0.370) 0.284 (0.351) 0.0152 (0.050) 1.738 (9.344) 4.389 (12.164) 0.667 (0.734) 8.974 (68.328) 0.101 (1.101) 0.406 (0.399) 0.144a (0.041) 1.146c (0.628)

Wind 1.457c (0.753) 0.207 (0.332) 0.239 (0.826) 0.844a (0.252) 0.120 (0.127) 0.0362 (0.089) 2.375 (1.777) 5.083 (6.506) 0.207 (0.309) 0.362 (0.404) 0.140 (0.618)

0.0340 (0.105) 0.0722 (0.833) 1.026 (6.097) 0.0447 (0.089) 5.880 (30.468) 0.0239 (0.359) 0.290c (0.151) 0.127 (0.140) 0.0416 (0.272) 0.146 (0.245) 0.125 (0.081) 0.375 (0.659) 0.0336a (0.006) 0.135a (0.033) 0.318 (0.202) 0.222 (0.176) 0.127 (0.166) 0.141 (0.119) 0.252 (4.031) 0.0861 (0.215) 0.378 (0.296)

Statistical signicance at .01 level. Statistical signicance at .05 level Statistical signicance at .10 level.

Table A2 Fixed effects from the ECM accounting for cross-sectional dependence. Aggregate renewable Australia Austria Belgium Canada Denmark France Germany Iceland Italy Japan 3.296 (8.83) 7.891 (26.29) 7.099 (61.07) 13.12 (10.90) 38.30 (24.75) 11.75 (50.94) 185.8 (274.02) 9.014 (9.05) 15.32 (18.63) 23.38 (15.97) Biomass 16.19 (95.64) 31.03 (28.34) 280.0 (204.52) 18.93a (5.32) 5.087 (19.68) 86.86b (41.71) 121.0a (31.51) 73.91c (43.22) 37.48a (8.61) Geothermal 2.512 (12.45) 14.87b (7.12) 7.940 (15.50) Hydro 13.01b (5.27) 8.450 (23.70) 12.35 (16.08) 11.45a (3.17) 34.75 (65.11) 6.510 (19.22) 37.69 (55.64) 1.176 (25.77) 7.570 (7.89) 16.25 (33.53) Solar 242.0a (93.34) 16.30 (256.94) Waste 169.0 (153.71) 77.74b (31.57) 12.73c (7.57) 40.78 (35.44) 78.30 (60.11) 1.617 (68.90) 72.14 (90.59) 104.9a (27.83) Wind 165.9 (212.89) 23.79 (41.34) 64.94 (160.88) 248.7 (425.71)

A. Ohler, I. Fetters / Energy Economics 43 (2014) 125139 Table A2 (continued) Aggregate renewable Luxembourg Netherlands New Zealand Norway Portugal Spain Sweden Switzerland United Kingdom United States Group Mean
a b c

139

Biomass 168.8 (295.38) 33.88b (17.12) 7.580 (15.99) 59.20a (22.93) 21.75 (109.09) 30.37 (20.45) 32.73 (38.52) 5.714 (15.59) 19.27b (9.28) 30.55 (20.48)

Geothermal 29.88b (12.45) 22.09a (5.88) 45.97 (30.47) 4.739 (11.03)

Hydro 1.531 (48.80) 253.5 (291.04) 30.98b (13.72) 14.43c (8.27) 54.10 (107.18) 57.53 (74.01) 64.52c (34.07) 5.779 (31.51) 11.36 (23.20) 35.27 (31.51) 13.55 (14.24)

Solar 48.49 (39.30) 90.02a (8.82) 15.85 (23.32) 15.08 (11.73) 30.98 (46.19)

Waste 4.011 (45.52) 78.37 (82.15) 192.7c (106.43) 46.55 (100.77) 24.46 (107.51) 4.770 (9.20) 37.80 (31.12) 20.29c (11.16) 13.15 (20.81)

Wind 12.81 (17.84) 93.67 (65.43) 60.18 (47.91) 49.95a (13.68) 358.2b (172.69) 34.67 (265.69) 96.68b (40.17)

44.55b (17.40) 121.9 (152.09) 47.37a (16.64) 7.477 (23.00) 85.85 (75.38) 51.62 (55.25) 47.04 (30.96) 58.64 (53.82) 12.89 (28.75) 54.55 (38.20) 10.24 (13.87)

Statistical signicance at .01 level. Statistical signicance at .05 level. Statistical signicance at .10 level. Lutkepohl, H., 1982. Non-causality due to omitted variables. J. Econ. 19, 267378. Marques, A.C., Fuinhas, J.A., 2012. Is renewable energy effective in promoting growth? Energy Policy 46, 434442. Menegaki, A.N., 2011. Growth and renewable energy in Europe: a random effect model with evidence for neutrality hypothesis. Energy Econ. 33, 257263. Ozturk, I., 2010. A literature survey on energygrowth nexus. Energy Policy 38, 340349. Payne, J.E., 2009. On the dynamics of energy consumption and output in the US. Appl. Energy 86, 575577. Payne, J.E., 2010a. A survey of the electricity consumptiongrowth literature. Appl. Energy 87, 723731. Payne, J.E., 2010b. Survey of the international evidence on the casual relationship between energy consumption and growth. J. Econ. Stud. 37 (1), 5395. Payne, J.E., 2011. On biomass energy consumption and real output in the US. Energy Sources Part B 6, 4752. Payne, J.E., 2012. The casual dynamics between US renewabl energy consumption, output, emissions, and oil prices. Energy Sources Part B 7 (4), 323330. Pedroni, P., 1999. Critical values for cointegration tests in heterogeneous panels with multiple regressors. Oxf. Bull. Econ. Stat. 61 (S1), 653670. Pedroni, P., 2001. Fully modied OLS for heterogeneous cointegrated panels. Adv. Econ. 15, 93130. Pedroni, P., 2004. Panel cointegration: asymptotic and nite sample properties of pooled time series tests with an application to the PPP hypothesis. Econometric Theory 597625. Perron, P., Vogelsang, T., 1992. Nonstationarity and level shifts with an application to purchasing power parity. J. Bus. Econ. Stat. 10, 301320. Pesaran, M.H., 2004. General Diagnostic Tests for Cross Section Dependence in PanelsInstitute for the Study of Labor (IZA). Pesaran, M.H., 2006. Estimation and inference in large heterogeneous panels with a multifactor error structure. Econometrica 74 (4), 9671012. Pesaran, M.H., 2007. A simple panel unit root test in the presence of crosssection dependence. J. Appl. Econ. 22 (2), 265312. Pesaran, M.H., Shin, Y., Smith, R.P., 1999. Pooled mean group estimation of dynamic heterogeneous panels. J. Am. Stat. Assoc. 94, 621634. Pirlogea, C., Cicea, C., 2012. Econometric perspective of the energy consumption and economic growth relation in European Union. Renew. Sust. Energ. Rev. 16, 57185726. Sadorsky, P., 2009a. Renewable energy consumption and income in emerging economies. Energy Policy 37, 40214028. Sadorsky, P., 2009b. Renewable energy consumption, CO2 emissions and oil prices in the G7 countries. Energy Econ. 31, 456462. Salim, R.A., Raq, S., 2012. Why do some emerging economies proactively accelerate the adoption of renewable energy? Energy Econ. 34 (4), 10511057. Westerlund, J., Edgerton, D.L., 2007. A panel bootstrap cointegration test. Econ. Lett. 97 (3), 185190.

References
Apergis, N., Payne, J.E., 2010a. Renewable energy consumption and economic growth: evidence from a panel of OECD countries. Energy Policy 38, 656660. Apergis, N., Payne, J.E., 2010b. Renewable energy consumption and growth in Eurasia. Energy Econ. 32, 13921397. Apergis, N., Payne, J.E., 2011a. A dynamic panel study of economic development and the electricity consumptiongrowth nexus. Energy Econ. 33, 770781. Apergis, N., Payne, J.E., 2011b. On the casual dynamics between renewable and nonrenewable energy consumption and economic growth in developed and developing countries. Energy Syst. 114. Apergis, N., Payne, J.E., 2011c. Renewable and non-renewable electricity consumption growth nexus: evidence from emerging market economies. Appl. Energy 88, 52265230. Apergis, N., Payne, J.E., 2011d. The renewable energy consumptiongrowth nexus in Central America. Appl. Energy 88, 343347. Apergis, N., Payne, J.E., 2012a. A global perspective on the renewable energy consumptiongrowth nexus. Energy Sources Part B 314322. Apergis, N., Payne, J.E., 2012b. Renewable and non-renewable energy consumption growth nexus: evidence from a panel error correction model. Energy Econ. 34, 733738. Apergis, N., Payne, J.E., Menyah, K., Wolde-Rufael, Y., 2010. On the causal dynamics between emissions, nuclear energy, renewable energy, and. Ecol. Econ. 69, 22552260. Binder, M., Offermanns, C.J., 2007. International Investment Positions and Exchange Rate Dyanmics: A Dynamic Panel AnalysisCenter for Financial Studies. Bowden, N., Payne, J.E., 2010. Sectoral analysis of the causal relationship between renewable and non-renewable energy consumption and real output in the US. Energy Sources Part B 400408. Breitung, J., 2001. The Local Power of Some Unit Root Tests for Panel Data, vol. 15. Emerald Group Publishing Limited. Breitung, J., Das, S., 2005. Panel unit root tests under cross-sectional dependence. Statistica Neerlandica 59 (4), 414433. Carrion-i-Silvestre, J.L., Barrio-Castro, T.D., Lopez-Bazo, E., 2005. Breaking the panels: an application to the GDP per capita. Econ. J. 8, 159175. Choi, I., 2001. Unit root tests for panel data. J. Int. Money Financ. 20 (2), 249272. Clemente, J., Montanes, A., Reyes, M., 1998. Testing for a unit root in variables with a double change in the mean. Econ. Lett. 59, 175182. Ewing, B.T., Sari, R., Soytas, U., 2008. The relationship between disaggregate energy consumption and industrial production in the United States: an ARDL approach. Energy Econ. 30, 23022313. Hadri, K., 2000. Testing for stationarity in heterogeneous panel data. Econ. J. 3 (1), 148161. Im, K.S., Pesaran, M.H., Shin, Y., 2003. Testing for unit roots in heterogeneous panels. J. Econom. 115 (1), 5374. Kao, C., Chiang, M.H., 2000. On the estimation and inference of a cointegrated regression in panel data. Center for Policy Research, Paper 145. Levin, A., Lin, C.-F., Chu, C.-S.J., 2002. Unit root tests in panel data: asymptotic and nitesample properties. J. Econom. 108 (1), 124.